What a Claude API gateway is, and when you need one

What a Claude API gateway actually does

A Claude API gateway is a service that speaks the Anthropic Messages API to your code and relays each request to the model upstream. Your tools point at the gateway exactly as they would point at api.anthropic.com; the gateway owns everything around the request — who is calling, what it costs, and which key is allowed to spend. You use one for practical reasons: a lower price, instant access without a provider account, or controls the provider does not offer.

• Presents the standard Anthropic Messages API, so SDKs and tools work unchanged.
• Authenticates your key and enforces per-key guardrails before anything goes upstream.
• Meters every request at official provider rates and handles billing — here, a prepaid balance at a flat 50% B2C discount.
• Records per-request usage with a token breakdown you can audit in a dashboard.

See also: apiToken.sale vs buying from Anthropic directly

Gateway vs proxy vs going direct

People use "gateway" and "proxy" interchangeably, and the difference matters when you pick one. A reverse proxy forwards bytes and understands nothing about them. A gateway understands the protocol and owns part of the request lifecycle — authentication, metering and settlement. Going direct means Anthropic does all of that for you, at official rates, with an Anthropic account.

The practical test: if deleting the middle layer changes nothing but the base URL, it was a proxy. If you would lose your billing, your limits and your usage history with it, it was a gateway.

How a request moves through the gateway

1. 01Your client sends a standard Messages API request to https://router.apitoken.sale/v1/messages with your key in the x-api-key header.
2. 02The gateway authenticates the key and checks its guardrails — the optional lifetime spending limit and expiration date — before routing anything upstream.
3. 03The model generates the answer. Streaming requests come back as standard Anthropic SSE events, token by token.
4. 04Usage is metered at official provider rates, your flat 50% B2C discount is subtracted, and the net amount draws down your prepaid balance.
5. 05The request shows up in your dashboard with its model and token-level breakdown, so spend is never a surprise.

Pointing an existing client at a gateway should mean changing two things: the base URL and the key. Keep your anthropic-version header and your model IDs. A service that demands a different request shape is a translation layer, not a native gateway — expect subtle breakage in streaming, tool use and prompt caching.

Native protocol, not a translation layer

apiToken.sale is Anthropic-native: any client that works against api.anthropic.com works against https://router.apitoken.sale/v1/messages, byte for byte. A minimal request looks exactly like the Anthropic docs say it should:

curl https://router.apitoken.sale/v1/messages \
  -H "x-api-key: sk-pool-•••" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

The same key is not limited to the Anthropic lane. It also speaks the OpenAI-compatible protocol at https://router.apitoken.sale/v1 with an Authorization: Bearer header, and the native Gemini protocol with x-goog-api-key — so one gateway key covers supported Claude, GPT, Gemini and Kimi models without a second account.

Where the 50% discount comes from

The gateway buys capacity in bulk and sells it from a prepaid, pooled balance. Each of your calls is first converted to official Anthropic spend — input, output, cache reads and writes metered separately — then the flat 50% B2C discount is subtracted, and only the net amount touches your balance. The balance never expires and there is no customer subscription, so idle time costs nothing.

Because metering mirrors the official rate card, the usual levers still work: cache reads are far cheaper than fresh input, and Haiku costs a fraction of Opus per token. The discount stacks on top of whatever your prompt engineering already saves.

Estimate a workload before you top up →

Per-model rates, context windows and cache pricing →

Key-level controls direct access does not give you

• An optional lifetime spending limit per key — a hard ceiling, useful for keys embedded in tools or handed to a team.
• An optional expiration date, so a temporary key dies on schedule instead of living forever in someone's config.
• Per-request usage visibility in the dashboard, broken down by model and token bucket.
• Instant issuance with no Anthropic account, waitlist or billing-country requirement — payable by bank card or cryptocurrency.

When you do not need a gateway

Be honest about the trade. If you already have frictionless Anthropic billing, an enterprise agreement, or a compliance requirement to contract with the model provider directly, going direct is the right call — a gateway adds a party to your request path, and that party has to be trustworthy. If what you want is the same models at half the official spend, instant onboarding, and prepaid predictability, a native gateway is the pragmatic choice.

Frequently asked questions

Related guides